Einstein Generative AI & Trust

Einstein Generative AI and Your Data

At Salesforce, trust is our #1 value. To keep your data secure, Salesforce has agreements in place with large language model (LLM) providers, such as OpenAI. These agreements allow organizations to use generative AI capabilities without their private data being retained by the LLM providers.

Trusted Generative AI

Salesforce’s Einstein generative AI solutions are designed, developed, and delivered based on our five principles for trusted generative AI.

• Accuracy: We back up model responses with explanations and sources whenever possible. We recommend that a human checks model responses before sharing with end users for the majority of use cases.
• Safety: We work to detect and mitigate bias, toxicity, and harmful responses from models used in our products through industry-leading detection and mitigation techniques.
• Transparency: We ensure that our models and features respect data provenance and are grounded in your data whenever possible.
• Empowerment: We believe our products should augment people’s capabilities and make them more efficient and purposeful in their work.
• Sustainability: We strive towards building right-sized models that prioritize accuracy and reducing our carbon footprint.

Reviewing Generative AI Responses

Generative AI is a tool that helps you be more creative, productive, and make smarter business decisions. This technology isn’t a replacement for human judgment. You’re ultimately responsible for any LLM-generated response you share with your customers. Whether text is human- or LLM-generated, your customers associate it with your organization’s brand and use it to make decisions. So it’s important to make sure that LLM-generated responses intended for external audiences are accurate and helpful, and that they align with your company’s values, voice, and tone.

When your end users review generated responses for external audiences, focus on the accuracy and safety of the content.

• Accuracy: Generative AI can sometimes “hallucinate”—fabricate responses that aren’t grounded in fact or existing sources. Before you publish a response, check to make sure that key details are correct. For example, is the customer service suggestion based on an existing and up-to-date knowledge article?
• Bias and Toxicity: Because AI is created by humans and trained on data created by humans, it can also contain bias against historically marginalized groups. Rarely, some responses can contain harmful language. Check your responses to make sure they’re appropriate for your customers.

If the response doesn’t meet your standards or your company’s business needs, you don’t have to use it. Some solutions allow end users to edit the response directly, and if not, it’s best to start over and generate another response.